Thank you guys for your replies. As always, it motivates me to push things further and tinker with my systems 😀
Today I obtained very nice Zalman VGA cooler - ZM80C-HP, so it was time to open up P4 machine. Unfortunately, it isn't latest revision with two heatpipes (ZM80D-HP), but it was brand new and cost me about 5 euro. I added 80mm fan (maybe I will find genuine Zalman unit one day) and it is more than enough for my Radeon 9700Pro. Maximum core clock went from 365MHz to about 425MHz! It's more than stock 9800XT! Earlier this week, I also got Zalman CNPS7700Cu for my CPU - stock cooling is for casual users 😁
Mounting ZM80C-HP was trickier than I expected, but I didn't mount clips sticking two heatsinks together, as it required a lot of force to push them in place, and made contact with GPU worse. To my surprise there wasn't any compatibility issues neither with chipset cooler nor RAM sticks (both are taller than usual). Actually RAM stick missed Zalman's heatpipe by about 1mm 😁
Unfortunately I managed to break my previous Radeon 9700 Pro, so now I am using AIW card. I think that it was baked by someone and sold, because it booted few times and then started to display complete mess. I didn't even installed drivers before it died. Anyway, for now I will settle with what I got, play with K6 and keep P4 under desk waiting for games which K6 cannot manage.
I also tested Antec VP-400 PSU which I recapped. It wasn't nicest job I've done, since basically any decent capacitor with required ratings is physically bigger than the crap used by CWT, but I managed anyway. Here are results (and that's P4 for you. CPU was at 3,45GHz 1,55V so imagine what would it look like with overclocked Prescott (keep in mind that I have Northwood. Pretty decent one)):

(from left to right: peak load (hovered around 296W most of the time). idle +12V voltage, load +12V voltage, load +12V volatage after 20 minutes of stress test)
I tested with ATITool 0.2.6 artifacts test set to highest priority in Task Manager and prime95 Small FFTs with single thread as I found that this configuration gives me highest power draw on this system. Antec manages this test nicely, but it runs very hot. I think I will swap it with something more recent and keep it for some Athlon XP build if I ever build one (lately I've got Barton Mobile 2500+ and Epox 8RDA3I so who knows 😀 ).
I also got very interesting card which is Gainward GeForce 7800GS in disguise. It has G71 chip with 24 pipelines and 512MB of 1,4ns RAM making it basically 7900GT. Of course with AGP slot. I chose not to plug it in with P4 as I think that it will fit AMD 939 build better, and I will build such PC one day as I very like this platform.
Here are current pics of my P4:

It isn't nicest looking system, and I even find it not being in my test by myself, but it has it's own charm and 2003 character 😀
I managed to get second Voodoo 2 which is Orchid Righteous Voodoo 2 12MB (thanks to tikoellner !) but I didn't found motivation to make SLI cable yet. Cards are of course unmatched badly with one being Diamond 8MB and the other Orchid 12MB, but maybe I will find second Orchid or second Diamond one day (Diamond 8MB will be much easier to find cheap tho).
As I am getting disappointed with K6-III+ performance, P4 surprised me positively. I didn't managed to run PAT with 1GB sticks, but thanks to tight RAM timings and high memory clock, performance is very nice for 478 build. Here are some numbers:

As always, looking forward to hearing from you guys. Let me know what you think about this build 😀
2017: 7800X@4,6G / X299 / 32GB / GTX 1080 / SM961 256GB+2x256GB RAID0 / G710+ / G402 / U2713H
2003: P4 2,8C@3,4G / IS7 / 2GB / AIW9700Pro / 160GB+2x40GB RAID0 / SK-8000 / IMO 1.1A / G200
2000: K6-3+@600M / 591P / 384MB / Voodoo3+1 / GUS+AWE32 / 40GB